Kev-0.5B

Quadrant scores

See the full quadrant

Scored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.

• Maturity5/10

  Open weights and a documented local System One–compatible API score well for access; no hosted SLA and a research footprint keep maturity mid-pack.

• Capability6.2/10

  Supports choice / score / boolean with usable laptop latency; author calib looks fine in-distribution, but transfer vs Jev is self-reported and weaker OOD.

• Adoption8/100

  Niche GitHub/HF interest among builders reconstructing Jev; almost no general awareness outside that circle.

Vendor claims

• ~160ms for a six-question request on Apple Silicon (laptop)[Vendor claim — not independently verified]
• In-distribution held-out accuracy ~0.799 / ECE ~0.065 (author eval)[Vendor claim — not independently verified]
• Transfer vs Jev ≈ −19pp on out-of-training suite (author comparison)[Vendor claim — not independently verified]

Kev is not a frontier model — it is a LoRA adapter + readout head on top of Qwen2.5-0.5B, an open reconstruction of the System One / Jev contract by Jared Palmer.

It reads a document once and answers many typed questions in parallel in a single prefill pass — no autoregressive decode. Question types: noul (yes/no → catalog boolean), choice, and score. Output is a calibrated distribution over options, not prose.

What it is

• Open weights: jaredpalmer/kev-0.5b (Apache-2.0 adapter; base under Qwen license)
• Repo + local server: github.com/jaredpalmer/kev
• API: POST /v1/systemone — TypeSafe-compatible shapes; official SDK works with a local base_url
• Architecture follows the independent reconstruction in Archer Hume — Jev’s Architecture Unmasked (17 Sep 2026): shared state, isolated question branches, pointer readout, proper scoring — not a TypeSafe confirmation

See also the catalog entry for the closed reference product: Jev.

Limits (read before you ship)

• Author label: research prototype — “not a production model and it is not Jev”
• Backbone is 0.5B; author reports ~−19pp transfer gap vs Jev off training distribution (self-reported)
• Residual option-order sensitivity (~7% argmax flips in author tests)
• Local serve, no auth — fine for laptop demos, not a hosted waitlist SLA
• Do not use for production decisions that affect people without your own labelled eval

Why it is in the catalog

Public typed decision I/O, documented System One-compatible API, and open weights. It sits next to Jev as an open / local point on the quadrant — with artifactKind: lora-adapter so the card does not pretend this is a frontier base model.